Using Image Recognition Technology and Smartphones to Improve Trichiasis Surgery Outcomes

摘要 沙眼性倒睫（TT）手术后的结果往往不理想， 倒睫（PTT）范围从10% - >50%，术后一年内，中度至重度眼睑 轮廓异常（ECA）发生率>10%。尼日利亚是世界上第二大 倒睫病例，目前有超过100，000人需要手术矫正。这些人生活在 在农村社区，获得医疗保健的机会有限。因此，手术是由眼科护士（倒睫 外科医生”），他们接受关于如何进行TT手术的简短课程，然后被要求进行手术 在他们地区的村庄里建立营地。往往缺乏支持性监督。在这个项目中，我们计划 开发移动健康工具，为卫生工作者提供关于手术的即时反馈， 提供关于是否需要额外手术调整的指导。 我们先前的工作已经证明，术后即刻的眼睑外观是一个强有力的预测， 术后成功;我们团队的外科医生能够预测70%的PTT病例和85%的眼睑 术后6周时的轮廓异常病例，基于闭合时每个眼睑的单张照片， 手术此外，我们的团队已经证明，我们可以开发图像识别软件来检测眼睑， TT与>90%的准确性（灵敏度92%）使用我们正在进行的NEI资助的临床试验的照片。在 在R21阶段，我们计划使用来自同一试验的图像来开发一台机器 可以预测眼睑是否将发展眼睑轮廓异常或PTT的学习算法 (AIM 1）。这种算法将被纳入一个智能手机应用程序，将允许外科医生在 手术结束后，立即收到关于眼睑是否可以从调整中受益的反馈 例如收紧或松开缝线（AIM 2）。在R33阶段，我们将测试 算法和应用程序使用迭代，以用户为中心的方法（AIM 3）。然后我们将与尼日利亚人合作 卫生部和合作伙伴开发和测试应用程序部署和手术监测协议（AIM 4）。最后，使用Aim 3的数据，我们将通过使用 经济模型来估计全球使用该应用程序可以避免的不良结果的数量 (AIM 5）。该项目有可能大大改善倒睫手术的结果在世界范围内。此外，本发明还 这个项目可以作为一个证明的原则，远程技术可以用来帮助外科医生和其他 在资源贫乏的环境中工作的卫生工作者。

ABSTRACT Outcomes following surgery for trachomatous trichiasis (TT) are often suboptimal, with rates of post-operative trichiasis (PTT) ranging from 10% - >50% within one year following surgery and moderate to severe eyelid contour abnormalities (ECA) occurring >10% of the time. Nigeria is home to the second largest number of trichiasis cases globally, with >100,000 individuals currently needing surgical correction. These individuals live in rural communities, with limited access to health care. Thus, surgery is provided by eye nurses (“trichiasis surgeons”) who receive a brief course on how to perform TT surgery and then are asked to conduct surgical camps in villages across their region. Supportive supervision is often lacking. In this project, we plan to develop an mHealth tool to provide health workers with immediate feedback on their surgery, and to provide guidance on whether additional surgical adjustment is warranted. Our prior work has demonstrated that the immediate post-operative eyelid appearance is a strong predictor of later post-operative success; a surgeon on our team was able to predict 70% of PTT cases and 85% of eyelid contour abnormality cases at six weeks post-op based on a single photograph of each eyelid at the close of surgery. Further, our team has shown that we can develop image recognition software to detect eyelids with TT with >90% accuracy (sensitivity 92%) using photographs from our ongoing NEI-funded clinical trial. In the proposed project, during the R21 phase, we plan to use images from the same trial to develop a machine learning algorithm that can predict whether an eyelid will develop either an eyelid contour abnormality or PTT (AIM 1). This algorithm will be incorporated into a smartphone app that will allow surgeons to take a picture at the end of surgery and receive immediate feedback on whether the eyelid could benefit from an adjustment such as tightening or loosening the sutures (AIM 2). In the R33 phase, we will test the functionality of the algorithm and app using an iterative, user-focused approach (AIM 3). Then we will work with the Nigerian Ministry of Health and partners to develop and test a protocol for app deployment and surgical monitoring (AIM 4). Finally, using data from Aim 3, we will assess the impact of the app on long-term service delivery by using economic models to estimate the number of poor outcomes that could be averted by using the app globally (AIM 5). This project has the potential to dramatically improve trichiasis surgery outcomes worldwide. Further, this project can be used as a proof-of principle that remote technology can be used to aid surgeons and other health workers operating in resource-poor settings.